[Bp_ipv6] Can we do the math? ( was [MEAC ICANN] Fwd: Interesting IPv6 metrics)

Omar de León - Teleconsult omar.deleon at teleconsult.world
Sun Aug 28 12:27:23 EDT 2016

Thanks folks for your information and considerations. 


You can find on the web site of LACNIC a report where economic considerations and evaluations are done for alternatives of transition to IPv6. Net Present Value is used as the quantitative tool to introduce time effect, technologies for transition evolution, Opportunity Cost of Capital (WACC), and other data, through an economic model. 

The report in English is at: http://portalipv6.lacnic.net/en/caf-lacnic/ Sections 5, 6 and 7 may be of your interest. 

The economic model is in Spanish but it would be translated into English in the near future. http://stats.labs.lacnic.net/PROYECTOCAF/modelo/


Reviews and comments will be welcomed.


About the bypass flux (v4 to v6) of applications and content when you provide IPv6 access it should be about 50% for most of the countries in the world (less Japan for example, maybe because of language). Please refer to page 39 of the report where it can be seen different indicators. The reason is that the most traffic generation sites are already on IPv6 as they are popular over the world. This means that the opening of IPv6 automatically discharge CGN and reduce the investment on mitigation techs.


Regards, Omar


De: Bp_ipv6 [mailto:bp_ipv6-bounces at intgovforum.org] En nombre de Sumon Ahmed Sabir
Enviado el: domingo, 28 de agosto de 2016 10:46
Para: Marco Hogewoning <marcoh at ripe.net>
CC: bp_ipv6 at intgovforum.org
Asunto: Re: [Bp_ipv6] Can we do the math? ( was [MEAC ICANN] Fwd: Interesting IPv6 metrics)


Many Thanks Marco for your elaborated notes. Fully agreed with you.


Regarding point 2) "How much is the cost difference between IPv4 and IPv6 transport?"


You have discussed NAT444 and NAT64/464XLAT. But in case of dual stack situation with NAT444 and native IPv6 its a clear case of cost savings on CGN part depending on how much traffic will shift from IPv4 to IPv6.  If we add the answer of question 3) that is around 50%. 


You mentioned its 50% or more in US. I have found the same result, its between 50% to 60% in a Asian developing country. Though I have worked on a very small sample.








On Fri, 26 Aug 2016 at 18:12 Marco Hogewoning <marcoh at ripe.net <mailto:marcoh at ripe.net> > wrote:

Hi Sumon,

Thanks. Regarding the issue on overbooking, I think for the more generic model that would be less of an issue. In using T-mobile I just picked a random example to see if the logic would lead to a reasonably realistic number. In fact, no need to calculate this for T-mobile, they are already doing IPv6.

In light of best practice, what we need to document is the model, as it would apply to a random ISP.

1) How much traffic do you ship?
2) How much is the cost difference between IPv4 and IPv6 transport?
3) How much traffic is expected to shift from 4 to 6 when it has the possibility?

The 3 numbers here will tell you your potential savings, which of course can also be plotted over time.

The answer to question 1 is really straightforward, every operator has that number in front of them all the time.

The second question needs a bit more work, but as an operator you should be able to piece it together. Of course it depends on the situation you are in and how much you pay for the various things that make up this number. In cases where CGN is an absolute must, the cost difference probably will be a bit higher than in those situations where native iPv4 is being used and dual-stack is a realistic scenario.

As a group, maybe we can add a bit of guidance here on how to realistically project or calculate this. Of course if I need to replace my core router to handle IPv6, that CAPEX will attribute to my IPv6 transport cost. At the same time, if you take into account traffic growth (answer 1 over time), it is probably safe to argue that the cost of a new router would have been there anyway and therefor does not contribute to the difference between the two protocols and just change the overall figure. At the same time, changing from NAT444 to an IPv6-only with NAT64/464XLAT (as T-mobile does), means that the cost for transporting the residual IPv4 traffic might be changing as well.

Hopefully we can find a few people, who maybe even without giving us real numbers, can provide some insight in how they did the math on this.

The other part where we, or more precise the people who already run IPv6, can contribute is question 3.

There are quite a bit of presentation materials and publications available that describe and quantify the amount of traffic that shifts from IPv6 to IPv4. And as you are planning to deploy IPv6, that is probably the only thing you have. You might be able to use some aggregated netflow data to predict what will happen and the exact shift depends on what your customers use.

We all know the big “content generators” (mind you, in reality it is your customer generating the data) are the Youtubes and Netflixes of this world, We also know that many of these providers offer IPv6: Google/Youtube, Netflix, Facebook all will be happy to see the traffic move. We also know a few that are just starting (Amazon) or still have to make the switch (a lot of the Microsoft cloud products).

Your mileage may vary, but under normal circumstances a shift of 50% as observed in the US is realistic. In fact, knowing the still have about a third on v4, the actual shift for a v6 customer should be higher than 50%

Maybe here we can collect some additional data and offer some table where people who do not have sophisticated measurements can get an idea on what to expect. Of course this data could also useful in planning and pilot projects to make sure the IPv6 handling part of the network is of sufficient scale to deal with the load.

And again, taking a step back, it is less about the actual number: Once figure out that half your traffic is Youtube/Netflix, that half can be expected to move.



> On 26 Aug 2016, at 08:22, Sumon Ahmed Sabir <sasabir at gmail.com <mailto:sasabir at gmail.com> > wrote:
> Excellent work Marco.
> Your maths are quite logical to me.
> Only major doubt in my mind is calculating throughput form the Average download speed 22.4 Mbps. Probably it does not mean people are using at a average of 22.4mbps all the time? Even though the 1:1000 ratio is very conservative, It seems to me that we can only make a realistic assumption if we get the average of total bandwidth consumption of T-Mobile.
> But again its seems a promising start, thanks again for the math.
> -sumon
> ২৫ আগস্ট, ২০১৬ ৬:১৯ pm এ, "Marco Hogewoning" <marcoh at ripe.net <mailto:marcoh at ripe.net> > লিখেছেন:
> (This may not reflect my employers opinion or even reality as a whole)
> All,
> Exec summary: 5 year ROI on IPv6 for a larger-than-life mobile carrier in the US can be gestimated at 225 million USD.
> Disclaimer: I might be off by a zero or two (factor 10/100) as a result of assumptions or flaky math :)
> Can’t find the original email on this thread, but somewhere Jan wrote:
> > I asked Cameron Byrne from T-Mobile USA how many mobile devices do they
> > have now on their network using IPv6 connectivity and his estimation was
> > around 48 million.
> >
> > Cheers, Jan
> Was chatting with Silvia the other day, wondering if we can piece it together….
> Curious mind wants to know:
> - How many customers are in this network?
> The answer (Q2 financial reporting) is 67,3 million (call it 67 for ease)
> We have them apparently also on record saying 50% of the traffic is IPv6. This doesn’t say much unless we know:
> - How much traffic is on that network?
> Now, this needs a bit of reading between the lines and guessing, but the Q2 investor report has one interesting bit where they brag on quality:
>         "In the second quarter of 2016, T-Mobile’s average 4G LTE download speed was 22.4 Mbps compared to Verizon at “
> So the average customer has 22 Mbps, and 67 million of them ((22*10^6) *( 67 * 10^6)) gives around 1,5 petabit/s, which is a rather big number
> In fact unlikely that all of them are pushing 22 mb _all_ of the time and I guess we can figure in some oversubscription. Conservative estimate of 1:1000 would bring the total average traffic down to 1,4 terabit/s, 50% of which is IPv6 :) (more likely the overbooking is a factor 10 higher, but let’s stick to this).
> - How much does this cost to transport that amount of bits?
> We don’t know the number for T-mobile, but we have a credible source in Swisscom:
>         http://www.swissipv6council.ch/sites/default/files/docs/map_martin_gysi_ipv6_council.pdf says difference is 6350 CHF.
> So, passing 1 gb/s through CGN costs 8300 USD, shipping the same amount of traffic on IPv6 is only 1700 (which btw is 5 times, cheaper, not 6 as often claimed by hearsay on social media).
> Putting this all together:
> If 50% of the traffic (0.7 tbit/s) really has moved, the cost saving in transport is in the order of 4.6 million USD per month at the current rate.
> Now of course this meant that a) the had a huge spend on IPv6 b) started with zero return. But still ever since they started to roll out, they have been collecting some money back from the first IPv6 bit shipped.
> I’m too stupid to figure out how to get Excel to model this across the past 36 months, but if I am not mistaking in lineair (which it isn’t) you end with 86 million.
> Last question remains:
> - How much did it cost them and is 36 months a reasonable ROI?
> Or can we extend to 5 years (at which point you are around 225 million USD).
> - Is 225 million a reasonable cost estimate to get a network with 70 million users to IPv6?
> MarcoH
> _______________________________________________
> Bp_ipv6 mailing list
> Bp_ipv6 at intgovforum.org <mailto:Bp_ipv6 at intgovforum.org> 
> http://intgovforum.org/mailman/listinfo/bp_ipv6_intgovforum.org

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://intgovforum.org/pipermail/bp_ipv6_intgovforum.org/attachments/20160828/17dda054/attachment.html>

More information about the Bp_ipv6 mailing list